EKELM-Based Cloud Resource Prediction with Butterfly Particle Swarm Brucker Optimization

Abstract

Cloud computing provides on-demand access to scalable computing resources. Accurate demand prediction and efficient resource allocation are important for reducing operational cost, energy consumption and SLA violations. In this paper, we propose an intelligent cloud resource management framework which combines Extended Kernel Extreme Learning Machine (EKELM) for predicting the resource demand and Butterfly Particle Swarm Brucker Optimisation (ButPSBO) for optimal allocation of the resources. EKEL learns nonlinear and dynamic workload patterns to predict future resource requirements. ButPSBO improves the allocation decisions by balancing exploration and exploitation with respect to SLA constraints. We assess the framework based on Cloud Sim-based simulated workload traces to model heterogeneous cloud task execution and resource-utilization patterns and compare it with traditional machine learning and optimisation techniques. The experimental results show that EKELM can achieve prediction accuracy of 93.2%, and the prediction error is reduced by 14.9% compared with ELM-based methods. Moreover, ButPSBO enhances resource utilisation by 17.8%, decreases energy consumption by 13.6%, and reduces operating cost by 15.2%. In summary, the proposed framework provides an intelligent, cost efficient, energy efficient and SLA aware resource management in dynamic cloud environments.